Sharpen a photo that is slightly soft

Blur is not a single thing. A picture can be soft because it was focused slightly in front of the subject, because the lens is soft at the edges, because it was enlarged, or because the camera moved. The first three spread each point into a small, round blob, and a small, round blob is what this stage can undo. The fourth smears in a line, and it cannot.

Richardson–Lucy repeatedly asks what sharp picture, blurred by a Gaussian of that width, would give the picture you have — and nudges its estimate toward it. Each step's correction is clamped, the final change is soft-limited, and the whole thing acts on brightness only, so it cannot draw colour fringes.

Over-deconvolution shows as ringing: faint repeated outlines beside strong edges, and a crunchy texture in flat areas. The loupe opens on the busiest patch at full resolution; if you see either, lower Strength or Softness.

• It cannot undo motion blur, camera shake or a badly missed focus. The information is not in the file.
• Faces and text that are genuinely blurred stay blurred; sharper-looking edges are not recovered detail.

Can it unblur any photo?

No. It sharpens slight softness — the kind a lens a little out, a picture resized down and up, or a gentle focus miss leaves. Motion blur and badly missed focus spread each point over many pixels in a way the file does not record, and no classical filter gathers that back.

How does it know how blurred my picture is?

It finds the strongest edges and measures how many pixels each takes to go from 10 to 90 per cent of its step. A Gaussian blur of deviation σ makes an edge about 2.56σ wide, so the width gives the blur to undo. You can move it if the result is too strong or too weak.

What happens when it detects motion blur?

It measures how one-sided the edge directions in the picture are. Motion smears edges along one direction, so the rest stand out; when that coherence passes 0.5, or the estimated blur is wider than 3 px, the page says the blur looks like motion and caps the strength at 40, because past that a classical deconvolution adds ringing, not detail.

Why does it look grainier afterwards?

Deconvolution amplifies fine detail, and grain is fine detail. The strength is already reduced on grainy frames; for a noisy picture, switch on noise reduction in the same panel first — it runs earlier in the chain.

Is this an AI deblur?

No. It is a classical, iterative deconvolution. A learned deblur model is not offered because none has been found that is commercially licensed, small enough to run in a browser and measured to work.

• Sharpen imageAmount, radius and threshold on brightness only, with a halo meter that says when edges start to glow.
• Image clarity enhancerClarity, sharpening and grain checked at 1:1 in a loupe on the busiest patch before you save.
• Denoise photoGrain measured first, colour noise removed, luminance grain reduced to taste, fine detail kept.
• Clarity and sharpening solve different problemsSeparate midtone local contrast from edge acutance before enhancing a photograph.